A style of panties that’s <s>pretty obscure and you’ve probably never heard of</s> characterized by it near fully covering the hips and stopping at their end while covering most or all of the butt. V2 made with AI Toolkit.<div data-type="blurb" data-id="73">You can help support making models like this <a target="_blank" rel="ugc" href="https://civarchive.com/articles/9229/how-to-give-a-model-creator-50-buzz-at-no-extra-cost-possibly-negative-to-you">by posting stuff you've made with it to the gallery using the "add post" button</a>, that helps me earn back the training cost at no cost to you.</div>Trigger: H_P_S_P Optional: [color] waistband (this actually works even without the LoRA!), crotch seam (same), Negative: lowleg panties, bow panties (you always need these for Illustrious panties), highleg (helps with certain checkpoints), photorealistic, photo (medium) (shouldn't be needed, but listed in case), seam, seamed legwear (stops vertical lines from appearing when crotch seam is used, also general problem with base model)Please tell me if the "Low Network" version works without issue or not. It was made with Network Dimension and Network Alpha set to 8 in training settings: This lowers file size (and thus RAM demands) substantially, but costs precision (but I have no idea if its relevant here given there's no inherent style). Since I'll be giving all these a V2 <a target="_blank" rel="ugc" href="https://civarchive.com/articles/35887/welp-training-with-any-r-images-will-cost-yellow-buzz-training-dot-civitai-only">before the window to do that closes</a> (assuming Toolkit with improved data isn't still worse than the comparatively poorly tagged Kohya original) I'm very interested in feedback on if this actually matters for this kind of LoRA. (Edit: From my initial testing the low network works far better, but still inferior to Kohya version.)Uses the revised dataset I used for Anima version with no changes.

Files with SHA256: a8725af2071ff742c6bc54c03265ffa8dc4b5f7b2ac73d973b3fb66f0597ad54